Cancer is a disease of considerable complexity. Cancer originates from and is supported by a wide variety of alterations in the genome that often lead to drastic alterations of the cell's control circuitry, producing a great deal of diversity in the molecular mechanisms operating in cancer cells. When attempting to make inferences from a diverse population, heterogeneity in either the network regulatory connections or operating rules blurs the relationships and rapidly reduces the ability to accurately discern regulatory interactions. This provides a considerable challenge to those attempting to determine which, of all the changes that can be seen, are consequential. However, knowledge of these regulatory changes will facilitate the discovery of prognostic markers and provide strong candidate drug targets. Current analytic methods rarely consider such complexity and heterogeneity. In addition, current methods to reverse-engineer regulatory mechanisms mostly concern a nearly homogeneous set of components that are nearly homogeneously co-regulated. This application proposes to develop computational methods that can search through heterogeneous sample sets to identify subsets of samples in which sets of genes that collaborate to carry out particular pathologic functions are homogeneously regulated. The method then utilizes identified subsets of samples with higher homogeneity to learn context-specific regulatory mechanisms. The method will use simultaneous analysis of a variety of molecular and clinical characterizations, and an analytic strategy that detects limited homogeneity of behavior against a background of high heterogeneity. The methods will be validated via subsequent analysis of combined genomic data, clinical information and available biological knowledge from three NIH funded projects which include multiple myeloma, glioblastoma and pancreatic cancers. Developed algorithms will be implemented as a set of computer software with graphical user-interface, which will be publicly available. [unreadable] [unreadable] This project is intended to produce types of analysis that are specifically designed to identify collaborative molecular behaviors in cancer via simultaneous analysis of multiple types of biomedical data, and to make them available to biomedical researchers, in the form of easily utilized software so that anyone with these types of biomedical data can use the methods developed in this project. Even a modest improvement in the ability to predict what patients would benefit from what treatments would significantly improve patient care. Understandings that would identify sets of synergistic drugs could have even higher impact. [unreadable] [unreadable] [unreadable] [unreadable]